Multi Label Classification for Emotion Analysis of Autism Spectrum Disorder Children using Deep Neural Networks

Abstract

Emotion recognition and analysis is the process of identifying emotions and feelings of a person. Emotion analysis process is accurate in identifying expression in normal people in a single attempt. Emotion analysis is difficult in case of Autism Spectrum Disorder (ASD) children which are suffering with communication problems and speech problems. This paper proposed an optimized deep learning model with multi label classification for predicting ASD and NoASD with emotion analysis in children of age group 1 to 10 years. The kaggle dataset [1] of 1857 ASD children and 1850 Typically Developed (TD) children are used in this paper. Proposed model performance is tested on Yale Expression Dataset [2], CAFÉ children dataset [3] and also tested on social media dataset of autism parents group. The model is implemented by extracting face landmarks and is used to predict ASD and NoASD as first classification label and emotion is detected based on landmarks by computing internal and external distances by feature wise. Convolutional Neural Networks (CNN) is used to work with extracted face landmarks by using optimization methods, dropout, batch normalization and parameter updating. The proposed model is applied to predict 6 emotions irrespective of 4 general emotions with better accuracy.